GKE ELEKTRONIK AB REPORT EDM, EMC, DRIVES Meters for ... meter report March 2010 1.pdfgke elektronik...

GKE ELEKTRONIK AB REPORT EDM, EMC, DRIVES GRANBERGSDAL, SWEDEN p. 1 INDUSTRIAL ELECTRONICS

GRANBERGSDAL 321 +46 586 12266 +46 70 651 07 84 [email protected] 691 92 GRANBERGSDAL, SWEDEN SKYPE: SKOGSGURRA www.gke.org

Meters for domestic kWh metering

Measurements on site, discussions, literature search and lab tests

February – March 2010

Summary

There is a suspicion that utility’s customer’s electricity meters have errors. Measurements on site do

not reveal any irregularities. Five meters have been tested in lab, three of the offending brand and

type and two other brands. The results show that all meters behave reasonably well under constant

and cycled load. When subject to cycled and thyristor (dimmer) controlled load, the suspect meters

show errors that are literally hundreds of percent (tables on page 12 and 13) while ‘brand B and C’

behave flawlessly with errors in the one – two percent range. Test set-up is available at GKE lab (can

also be moved) for verification or extended tests.

Background

One of the utility’s customers has observed that his energy bill seems to be on the high side. A

modest estimation from his side is that he is paying for around 15 percent more energy than he

actually uses. The utility has done extensive tests with double and triple meters, that have verified

the customer’s observations. They also hired an external consultant, S-E Berglund Elkonsult AB, to

find out if there are any anomalies in grid or installation. Nothing spectacular was found. Meter

supplier has also made measurements on site, but no results from that investigation are available to

the utility. When asked, the supplier only mentions that he “Found something”.

There was a suspicion that high-frequency phenomena or extraordinary aberrations were present

and GKE was contacted to find out if that was the case. Measurements on site were made Thursday

February 11, 2010. This report is about findings from that occasion as well as observations made

during discussions and lab measurements made by GKE.

Findings on site

Several measurements were made with different loads activated. No high-frequency phenomena

were found and no exceptional distortion. One characteristic load is a 2 kW heater connected to 400

V, It is cycled on/off quite rapidly. A recording is shown in picture 1 (visual aliasing present).



Picture 1. Heater cycled on for 8 seconds and off for 7 seconds (CI, CII and CIII)

A zoomed (around 250x horizontally) part of picture 1 shown in picture 2. Picture 1 shows only one

current for less clutter, picture 2 shows all channels.

Picture 2. Voltages and currents zoomed. Metering pulse also shown.

Currents I2 and I3 are obviously one and the same, flowing between U2 and U3 (400 V connection).

U2 has an additional switch mode supply connected, the little hump in I2 coincides with U2 peak,

which is a sure sign. U1 does not have much connected, and the little there is seems to be a linear

load that draws a sinewave with inductive phase-angle.



Discussion after visit on site

When looking through the recordings, one unexpected feature was noted – the fact that the

metering pulses seemed to be delayed with regard to actual power. There must always be a delay,

producing a pulse ‘in advance’ is impossible, but the delays observed were longer than expected.

Furthermore, there seemed to be a bias with regard to power cycling so that delay was less at the

beginning of a high power cycle than at the end. On several occasions, there was also an extra pulse

following close to the ‘ordinary’ pulse after power was cycled back to low. Such a bias could be an

explanation to the observed differences between recorded energy and expected energy.

It was decided that GKE should have a closer look at the power/pulse frequency relation and see if

there was a systematic error in the meter type.

Tests at GKE lab

It was soon realized that a wider variation in cycle duration and duty cycle was needed in order to

see any significant and consistent differences between real energy and energy recorded by the

meter. The utility sent a meter to GKE and a set of tests were carried out. The test plan included:

1. Calibration of transducers and recording equipment with constant power

2. Building a pulsed load that could be adjusted with regard to power levels and on/off cycling

3. Design and test of an Excel spreadsheet that would allow close examination of recorded

energy vs actual energy

4. Verification that transducers and recording equipment errors were within acceptable limits

also when load was phase controlled (thyristor control) and cycled

5. Running the actual tests with load in a range from 35 W up to near 3 kW and cycle times

from half a second to tens of seconds.

6. Analysis and, if necessary, rechecking results

7. Report

Transducers and recording equipment calibration

Even if the task is not an audit round, it is important to have all tools in good order. So, a rough test

and calibration was done. It is described in appendix 1.

The tests and calibrations resulted in a couple of changes in the original plans: Introduction of two

power levels instead of switching one power level on/off and substituting shunt resistors for current

clamps.

The two power levels were necessary in order to avoid cold start inrush current in the incandescent

lamps in the load bank. Starting current up to four times rated current was recorded. Such high peaks

consume at least two bits of resolution in the A/D converter and by using a preheat power level, the



inrush current could be reduced to insignificant levels. Also, it is necessary to have a certain power

flowing through the meter in order to maintain pulse output. The Ws pulse goes constantly true if

power is zero and that disturbs the measurements.

AC current clamp transformers work well for continuously varying currents. But with thyristor

controlled currents, there is a disturbing DC level when primary current is zero. Rather than

discussing if that is tolerable or not, we decided to use shunt resistors with low inductivity instead.

The nominal scale factor is 1 V/A. Small adjustments for actual resistance are made in the

spreadsheet for power calculation, but not in the ARCUS recordings.

Test set up

Picture 3. Test set up, simplified diagram. Auxiliary power, N and PE connections not shown.

The test rig is an ad hoc collection of bits and pieces available in the lab. The control unit (lower left)

is a function generator with settable cycle time and duty cycle. It controls a relay that connects a

potentiometer (PWR HI) to the control input of the 400 V, 35 A thyristor controller. The low power

level is set with another potentiometer (PWR LO).

The load bank is a set of incandescent lamps that are used for various tests and are very well suited

also for this test. One more load bank is available (with 12x300 W) for a total load of around 5.5 kW.

Lamps can easily be unscrewed for different loads (in addition to setting levels with the control unit)

and also for unsymmetrical loads.

Heavy duty alligator clips connect the load bank to the thyristor controller output. It is possible to

connect them to the thyristor controller input for ‘cleanest’ possible waveforms. Load levels then

have to be controlled by screwing lamps in and out.



The ARCUS has ten channels available. Channel 6 is used for metering pulses, which are picked up

from the LED on front of the meter under test. A dual pulse counting and comparison feature is

available, but is not used in this application. ARCUS ‘talks’ to the PC over a USB link and data are

streamed directly to disk. Data can then be used in Matlab or Excel. The ARCUS software has

generous cursor measurement possibilities, there is also a possibility to define ‘palettes’ with

measurements crafted to different applications. A presentation and a data sheet are available in

appendix 2.

The Excel sheet

The rather high sampling rate of the ARCUS results in a massive amount of data. A 60 seconds

recording produces close to 200 000 lines with ten columns in the Excel sheet. The tight sampling and

good accuracy makes accurate calculation of power and energy, also with very distorted waveforms,

possible. The main error sources are the voltage and current transducers, which can be calibrated

against known good instruments see appendix 1. There is also a possible source of inaccuracy in the

exact determination of the sample rate that will affect energy calculations. That is also discussed in

appendix 1.

A sample Excel sheet is shown in picture 4.

Picture 4. Excel sheet.



The example Excel sheet shows a ‘sanity check’ made with around 2700 W power level. Columns A –

F contain voltages and currents. The values shown are in millivolts and the scale factors used are

0.041 V/mV and 0.001 A/mV.

Column G calculates instantaneous power using the formula shown in box H6. The resulting graph is

shown as Serie 1. This graph should be a straight line (power is constant in a three-phase sine wave

system), but it is not. The reason is that we a) have some distortion in the waveforms and b) that the

first part of each current cycle is cut out by the thyristors. Picture 5 shows an example of the

waveforms resulting in the data in picture 4. The yellow box “P_medel” is the mean power in column

G. It corresponds quite well, error is less than 2 percent, to the power measured with the Norma

class 0.2 electrodynamic power meter. Proof that the sanity check worked out well.

Picture 5. Showing power being switched from high to low level.

Picture 5 shows maximum power during the first 400 milliseconds, and thereafter minimum power

level. The different traces can be identified using the international resistor colour code (0=black,

1=brown, 2 is red etcetera).

It is interesting to note that the current peaks are low just after the switch from high to low power.

The reason is an ‘inverse inrush effect’ where the filaments are quite hot at first and then cool down

so that resistance decreases and current peaks increase.

That fact that the first part of the sine wave (current) is missing also at maximum power level results

in the variations of total power shown in the graph in picture 4.



Lab tests

First run, one meter tested

The test set-up shown in picture 3 was used for preliminary tests of one of the offending meters. S0

pulses were picked up optically from the LED on front of the meter and the blink rate was compared

to recordings made with the ARCUS and a modified program.

The meter behaved well as long as the load was constant and linear. Cycled load and phase

controlled load resulted in large deviations between meter perception of power and what was

measured with ARCUS/Excel and also indicated by the analogue meter.

Discussions with officers at the utility led to a decision to test not just one, but several, of the

offending meters. Two other meter brands should also be tested and comparison should be made

within the group of meters as well as to actual power and energy measured by the ARCUS/Excel

system and the analogue class 0.2 electrodynamic meter. So, we were now testing five meters in all.

The meters will be identified as #1, 2, 3, 4 and 5 in the following tables and discussions. #1 is the

original meter, #2 and 3 are same brand and type while #4 and 5 are other brands.

The original plan, to use the S0 outputs for meter pulses, did not work. Meters #1 – 3 have coded

output signals and not the simple Ws/pulse relation found in the panel LEDs. Meters #4 and 5 have

the simple Ws/pulse relation on their S0 outputs. To resolve the situation, a set of improved optical

pick-ups was fabricated and used for all meters. Picture 5 shows part of the test set-up. The optical

pick-ups (black/yellow wires) are seen attached to the front of the meters.

Picture 5. Test set-up.



All meters were connected in series and then, via a thyristor module, to the load bank seen to the

left. There is also a power relay parallel to the thyristor module for direct switching without any

phase delay. Thyristor module and relay are controlled by the function generator situated between

meters and analogue instrument. A set of shunt resistors and their cooling fan can be seen just to the

right of the function generator. The ARCUS data acquisition unit is to the extreme right in picture 5.

Test variables

The only variable that isn’t changed during the tests is the mains voltage and its frequency. It remains

at its incoming value and is not controlled in any way. All other variables can be changed within wide

limits. Table 1 shows the possible variations. All domestic load variations can be run, except motor

loads, computers, TV sets, low energy lamps and other similar equipment. They can, however, be

added to the load bank if needed.

Variable type Controlled by Limits Remark

Total Load # of lamps connected 0 – 2925 W Nominal power at 400 V

Thyristor phase angle 10 – 180 degrees 0 – 2850 W

Load period Relay 0.5 – 10 seconds

Thyristors 0.1 – 10 seconds

Duty Cycle Relay 10 – 90 % Only when period > 1 s

Thyristors 10 – 90 % Over full range

Load imbalance # of lamps connected 0 – 100 % 100 % when all lamps in one

phase are disconnected

Phase angle Thyristors 0 – 170° Delay in current on

“True” phase

angle

Re-arranging load

between phases

None, but restricted

to certain angles

Phase angle changed in steps

by connecting “wrong way”

Distortion Thyristors 0 – ‘very high’ Highest distortion at 170°

Table 1. Load characteristics and limits.

The number of possible combinations is close to infinite, so a set of variable combinations was

decided on. The tests are presented in the order they were carried out.

Test method

All meters are connected in series and to one and same mains voltage. Different loads were run and

all data were logged with an ARCUS recorder. Six channels record voltages and currents and four

channels record Ws-pulses output via the LED on the meter’s front panel. Time between number of

pulses and number of pulses were counted from the recordings and corresponding power level

calculated. U and I recordings are used to calculate power in Excel. Picture 6 shows the principle.



Picture 6. Showing principle for pulse evaluation in ARCUS.

The number of pulses shown is five. The number of intervals between pulses is four. Each pulse

interval represents 3600 watt-seconds (or joule) and by dividing total energy (in this case 4x3600 Ws)

by total time, an accurate measure of the power – as perceived by the meter – over the interval can

be calculated. In this example, the power is 4x3600/5.129 = 2808 W.

ARCUS can accept up to four Ws pulse channels and it is thus possible to compare four meters under

identical conditions. Comparison can be made against a ‘known-good’ meter or against power

calculated from ARCUS collected voltage and current data.

The power metering function in ARCUS is based on the samples collected during the test and a

typical 30 seconds run results in around 100 000 data points for each channel. Or, in other words, 70

samples per mains cycle (at 50 Hz). It is, therefore, possible to measure true power also when

waveforms are badly distorted or severely cut by thyristor phase control.

The power calculation can be done in Excel, but a ‘pulse_search_and_Power_calculator’ module is

available in ARCUS for computers that do not have Office installed or when the simplicity of working

within one program environment is preferred over the ‘bells and whistles’ that Excel excels in.

First pulse Last pulse

Cursor I

Cursor II

Time between

cursors



Test with continuous load, no switching, no thyristors

Four meters were tested in a test run with direct on line connection and power level selected by

screwing lamps out or in the load bank.

Three power levels were used with all lamps available (3x975 W), medium level (2x300 W) and low

level (3x75 W). All power levels are nominal, the exact power is dependent on mains voltage and the

actual power consumed by the lamps.

Continuous power, no switching, no thyristors

Maximum power, 60 seconds recording.

Meter Reading 1 Reading 2 Power Remarks

1 46 pulses 58.989 s 2807 W

2 46 pulses 58.989 s 2807 W

3

4 46 pulses 59.188 s 2798 W

5 20 pulses 2.578 3161 W 10 000 pulses/kWh HIGH

W-meter

ARCUS 2826 W

Medium power, 60 seconds recording.


1 20 pulses 42.274 s 1703 W

2 22 pulses 44.325 s 1787 W

3

4 20 pulses 42.319 s 1701

5

W-meter

ARCUS 1791 W

Low power, 60 seconds recording.


1 3 pulses 46.244 s 234 W

2 3 pulses 46.429 s 233 W

3

4

5

W-meter

ARCUS 239 W



Looks just about right. Meter 5 high in one case. Meters 1 and 2 lower than meter 5 and ARCUS in

one case. More on meter 5 later.

Test with on/off switching using power relay – no thyristors

Maximum power, 60 seconds recording, 1 Hz, 50 % Duty Cycle


1 17 pulses 38.925 s 1572 W Double checked

2 15 pulses 36.704 s 1667 W Double checked

3

4

5

W-meter

ARCUS 1531 W Inrush current clipped on negative side

Medium power, 60 seconds recording, 1 Hz, 50 % Duty Cycle


1 11 pulses 41.707 s 949 W

2 11 pulses 41.665 s 950 W

3

4

5

W-meter

ARCUS 966 W

Low power, 60 seconds recording, 1 Hz, 50 % Duty Cycle


1 1 pulse 29.572 s 122 W

2 1 pulse 29.980 s 120 W

3

4

5

W-meter

ARCUS 126 W

Power cycling between 0 and nominal power cannot be done with too low a frequency. Most meters

switch off below a certain current level. Cycling at 1 Hz is safe and does not switch meters off.

Power cycling was believed to one of the reasons for meter errors. But it does not seem to influence

meters significantly. At least not when switching at 1 Hz. Next test is with thyristor controlled

switching. The customer’s load seemed to be thyristor controlled.



Pulsed load, thyristor controlled switching

Nominal maximum power 2700 W, 1 Hz, 50 % Duty Cycle


1 13 pulses 28.289 s 1654 W HIGH

2 13 pulses 28.442 s 1645 W HIGH

3

4 11 pulses 27.664 s 1431 W OK

5 20 pulses 4.892 1472 W 10 000 pulses/kWh OK

W-meter

ARCUS 1470 W



1 10 pulses 26.795 s 1344 W VERY HIGH


3

4 3 pulses 21.205 s 509 OK

5 20 pulses 13.804 s 522 W OK

W-meter

ARCUS 534 W





3

4 0 pulses Not enough power to produce one pulse in 30 s

5 10 pulses 20.404 s 176 360 Ws/pulse OK

W-meter

ARCUS 188 W

Cycling load with thyristors produces large errors. So large, in effect, that a very painstaking search

for all possible sources of error was conducted. Alternative measurement equipment and

transducers were used and detailed scrutiny of waveforms, math, everything was undertaken. No

errors were found. Also, the fact that meters #4 and 5 and ARCUS agreed closely was taken as an

indication that meters #1 and 2 were behaving very erratic. A test with continuous (not cycled)

thyristor controlled load was done. Results on next page.



Continuous, level controlled with thyristor firing delay

Maximum power, 30 seconds recording.


1 16 pulses 27.351 s 2106 W LOW

2 16 pulses 27.436 s 2099 W LOW

3

4 20 pulses 28.612 s 2516 W OK

5 20 pulses 2.869 s 2509 W 10 000 pulses/kWh OK

W-meter 2496 W 3x832 W

ARCUS 2556 W

Medium power, 30 seconds recording.




3

4 5 pulses 24.480 s 735 W OK

5 20 pulses 9.763 s 737 W 10 000 pulses/kWh OK

W-meter 645 W 215 W on L1, uneven currents – L3 higher

ARCUS 756 W

Low power, 30 seconds recording.


1 11 pulses 27.758 s 1427 W Triple checked EXTREMELY HIGH

2 11 pulses 27.634 s 1433 W Triple checked EXTREMELY HIGH

3

4 1 pulse 16.253 s 221 W OK

5 16 pulses 26.454 218 W 10 000 pulses/kWh OK

W-meter 222 W 3x74 W

ARCUS 233 W

The algorithms in meters #1, 2 and 3 (which are all of the same brand and type) seem to have big

problems with non-sinusoidal currents. It looks like it is not the load cycling per se that is causing the

meter errors, but the current’s waveform.

Finally, a test with load connected between L1 and L3 was carried out. This is to separate the error

mechanisms that produce the large error in meters # 1 – 3. If the error is an inability to take care of

phase angles other than zero (a rather unlikely assumption, but easily checked) then the extra 30

degrees should produce a corresponding error. Results in the following table.



Test with load connected L1 – L3 instead of L1, L2, L3 – N.


1 17 pulses 54.691 s 1119 W OK

2 17 pulses 54.723 s 1119 W OK

3 17 pulses 54.823 1112 W OK

4 18 pulse 54.823 s 1116 W OK

5

W-meter 1110 W

ARCUS

All meters behave as expected. Also when extra 30 degrees of phase shift introduced.

The conclusion is that meters #1 – 3 seem to have a “smart” algorithm that somehow checks for

current position with regard to voltage and uses that information to calculate power. That strategy

may work quite well for currents where the distance between zero crossings is one half cycle. But,

the resulting power when current is cycled with less than 180 degree conduction angle is not only

questionable – it is totally and completely wrong.

Meters #4 and 5 do not try to be smart. They sample often enough to measure current and voltage

correctly and multiply instantaneous values to arrive at correct power and energy.

It is time to talk to the meter manufacturer about this. The test set-up will be kept for a while if more

measurements need to be made.

Granbergsdal, March 8, 2010

Gunnar Englund



APPENDIX 1

CALIBRATION OF ARCUS AND TRANSDUCERS

Current transducers

The Chauvin-Arnoux Mini 05 current clamps that were used on site are AC clamps and there is a

problem when thyristors are switched off. The magnetic field in the clamp core collapses when the

continuous current path is interrupted by the thyristor (similar to inductive kickback, but milder) and

induces a voltage in the secondary coil. The voltage will influence the calculated power and introduce

a considerable error in the measurements. It was therefore decided not to use the Mini 05 and work

with calibrated shunt resistors instead. Picture A1.1 illustrates the problem with decaying field.

Picture A1.1 False signals when thyristors gated off.

The shunt resistors were constructed from standard 1 ohm, 4 W wire-wound power resistors in a

series-parallel configuration so that four resistors resulted in a 16 W resistor with almost zero

percent resulting error. Unfortunately, there were not enough resistors with the required plus and

minus tolerances to produce more than one zero error combination. The two other combinations

therefore have ‘normal’ deviations from nominal value. By calibrating the three shunt resistors

against a known current and in their intended channel allocation, we had the double benefit of

having traceability ‘from current to screen’. Picture A1.2 shows calibration and resulting scale actors.



Picture A1.2 Shunt resistor calibration and resulting scale factors.

The scale factors are close enough to unity and the resistors were used without any correction in the

‘sanity check’ measurement described in the section about Excel sheet. It should also be noted that

the power reading was also a little rough. The purpose of the check was simply to check that all parts

were doing their work properly.

The resistors are used in the same phases as they have been numbered above. See table A1.1.

Phase Scale Factor [A/V]

L1 1.0025

L2 1.0338

L3 1.0211

Table A1.1 Shunt resistor scale factors

Voltage divider

Arcus has a basic input range from -10 V to +30 V. That range has been selected for several reasons,

the recorder is designed for simplicity and reliability and also for use in industrial applications where

analogue signals in the +/-10 V range and control signals usually are between 0 – 12 or 24 V. The



chosen input range is adapted to such signals. The fact that process signals (4-20 mA) usually develop

a voltage drop between 5 and 10 V, full range across receivers and that all input pairs can be

configured to measure differential signals in that range with good common-mode rejection makes

the ARCUS an ideal recorder for industrial use. For other signals, adapters have to be used.

One range that is easily adapted by means of passive voltage dividers is the mains voltage. The

voltage divider used in these measurements was originally used as an ‘artificial zero’ for

measurements on VFDs. Adding three 1 kohm resistors makes it useful as a simple three-phase

voltage divider. Calibration can be seen in picture A1.3:

Picture A1.3 Voltage divider calibration and scale factors

Phase Scale Factor [V/V]

L1 40.425

L2 40.425

L3 39.945

Table A1.2 Voltage divider scale factors



ARCUS calibration

The ARCUS analogue front end does not contain any amplifiers. That makes the device very stable,

the scale factor is determined by metal film resistors and the A/D converter’s reference voltage. Both

are stable with time and temperature so that, once a calibration has been carried out, the scale

factors remain for long time and over a wide temperature range.

Calibration is a simple procedure that utilizes the “arcus_setup_and_calibration.exe” program.

Running it produces the following screens:

Identification of ARCUS unit:

Zero out

offsets:

Scale Factor screen finds and stores

gains for each channel.

The calibration procedure takes less than five minutes and results in an overall accuracy better than

+/-0.2 percent of range. Offsets and gains are stored in non-volatile memory on-board the ARCUS

device and is not dependent on any data stored in the PC.

Date post:	16-May-2020
Category:	Documents
Upload:	others
View:	4 times
Download:	0 times

GKE ELEKTRONIK AB REPORT EDM, EMC, DRIVES Meters for ... meter report March 2010 1.pdfgke elektronik...

Documents